Learning-Based Thermal Estimation in Multicore Architecture

1. A computer-implemented method, comprising: receiving an application to run on a hardware processor, the application having unknown temperature data, the hardware processor comprising a plurality of cores; obtaining hardware resource utilization data associated with the application, wherein the hardware resource utilization data is obtained at least by dynamically running a region of the received application for a first percentage of predetermined amount of time and using the obtained hardware resource utilization data as being applicable to rest of the predetermined amount of time, the dynamically running performing periodically to capture code phase changes, the hardware resource utilization data including at least a fan speed associated with running the received application and a cooling liquid flow associated with running the received application, wherein temperature data associated with running the application is to be automatically predicted; executing a trained neural network with the hardware resource utilization data associated with the application, the trained neural network predicting core temperature associated with running the application on a core of the hardware processor, wherein the trained neural network parameterizes at least fan speed and cooling liquid flow associated with the hardware processor, wherein the neural network is trained to, given an input including previously unseen hardware resource utilization data associated with an individual application, predict the individual application's core temperature associated with running the individual application; and based on the core temperature predicted by the trained neural network, controlling the plurality of cores to run selective tasks associated with the application.

2. The method of claim 1 , wherein the controlling comprises distributing tasks associated with the application across the plurality of cores to run, based on the core temperature.

3. The method of claim 1 , wherein the obtaining comprises receiving historical usage data associated with the application.

4. The method of claim 1 , further comprising pre-processing the hardware resource utilization data and wherein the trained neural network is executed with the pre-processed hardware resource utilization data.

5. The method of claim 1 , wherein a neural network is trained based on hardware resource utilization data associated with a plurality of applications, to produce the trained neural network.

6. The method of claim 5 , wherein the plurality of applications is profiled to determine the hardware resource utilization data associated with a plurality of applications.

7. The method of claim 1 , wherein the hardware resource utilization data comprises data associated with usage of at least one of a hardware core, a hardware memory, a fan and a liquid flow component.

8. A computer readable storage medium storing a program of instructions executable by a machine to perform a method comprising: receiving hardware resource utilization data and corresponding core temperature data associated with running a plurality of applications, wherein the hardware resource utilization data is obtained at least by dynamically running a region of the received application for a first percentage of predetermined amount of time and using the obtained hardware resource utilization data as being applicable to rest of the predetermined amount of time, the dynamically running performing periodically to capture code phase changes; and based on the hardware resource utilization data and corresponding core temperature data, training a machine to predict a future core temperature given usage data of an input application with unknown temperature data to run on a hardware processor, the training comprising building a neural network according to configured hyperparameters and allowing the neural network to learn autonomously to predict the future core temperature based on the hardware resource utilization data and corresponding core temperature data, wherein the neural network parameterizes at least fan speed and cooling liquid flow associated with the hardware processor, wherein the usage data of the input application includes at least the fan speed and the cooling liquid flow associated with running the input application, wherein the neural network is trained to, given an input including previously unseen hardware resource utilization data associated with an individual application, predict the individual application's core temperature associated with running the individual application.

9. The computer readable storage medium of claim 8 , further comprising profiling each of the plurality of applications running on at least one hardware processor to determine the hardware resource utilization data and corresponding core temperature data.

10. The computer readable storage medium of claim 9 , further comprising storing the hardware resource utilization data and corresponding core temperature data in a database of application profiles.

11. The computer readable storage medium of claim 8 , further comprising: determining hardware resource usage associated with running the input application; running the neural network with data associated with the hardware resource usage as input, the neural network outputting the future core temperature associated with running the input application.

12. The computer readable storage medium of claim 11 , further comprising: based on the future core temperature output by the neural network, distributing tasks associated with the input application to a plurality of cores associated with the hardware processor, the plurality of cores executing the tasks respectively.

13. A system comprising: at least one hardware processor; and a memory device coupled to the at least one hardware processor; the hardware processor operable to at least: receive an application to run on a target hardware processor, the application having unknown temperature data, the target hardware processor comprising a plurality of cores; obtain hardware resource utilization data associated with the application, wherein the hardware resource utilization data is obtained at least by dynamically running a region of the received application for a first percentage of predetermined amount of time and using the obtained hardware resource utilization data as being applicable to rest of the predetermined amount of time, the dynamically running performing periodically to capture code phase changes, the hardware resource utilization data including at least a fan speed associated with running the received application and a cooling liquid flow associated with running the received application, wherein temperature data associated with running the application is to be automatically predicted; execute a trained neural network with the hardware resource utilization data associated with the application, the trained neural network predicting core temperature associated with running the application on a core of the target hardware processor, wherein the trained neural network parameterizes at least fan speed and cooling liquid flow associated with the hardware processor, wherein the neural network is trained to, given an input including previously unseen hardware resource utilization data associated with an individual application, predict the individual application's core temperature associated with running the individual application; and based on the core temperature predicted by the trained neural network, control the plurality of cores to run selective tasks associated with the application.

14. The system of claim 13 , wherein the at least one hardware processor is operable to control the plurality of cores by distributing tasks associated with the application across the plurality of cores to run, based on the core temperature.

15. The system of claim 13 , wherein the at least one hardware processor obtains hardware resource utilization data associated with the application by receiving historical usage data associated with the application.

16. The system of claim 13 , wherein the at least one hardware processor further pre-processes the hardware resource utilization data and wherein the trained neural network is executed with the pre-processed hardware resource utilization data.

17. The system of claim 13 , wherein a neural network is trained based on hardware resource utilization data associated with a plurality of applications, to produce the trained neural network.

18. The system of claim 17 , wherein the plurality of applications is profiled to determine the hardware resource utilization data associated with a plurality of applications.

19. The system of claim 13 , wherein the hardware resource utilization data comprises data associated with usage of at least one of a hardware core, a hardware memory, a fan and a liquid cooling component.